feat(kernel): richer TLS Go options — CA bundle + independent hostname skip#400
feat(kernel): richer TLS Go options — CA bundle + independent hostname skip#400mani-mathur-arch wants to merge 1 commit into
Conversation
543c7a2 to
ba466d3
Compare
96ba29c to
ab90eff
Compare
ba466d3 to
f35b387
Compare
ab90eff to
0abfb38
Compare
f35b387 to
d0c999e
Compare
0abfb38 to
89c9b36
Compare
…e skip
Expose the two kernel-only TLS knobs whose C-ABI setters already exist on
kernel main (no kernel change needed), via an experimental-option idiom:
- WithKernelTrustedCerts(pem) -> kernel_session_config_set_tls_trusted_certs,
adding a PEM CA bundle on top of the system roots. Required because the
kernel's rustls stack ignores SSL_CERT_FILE, so a custom CA (corporate
re-signing proxy / on-prem CA) must be handed over explicitly.
- WithKernelSkipHostnameVerify() -> set_tls_skip_hostname_verification,
skipping only the hostname check while keeping chain validation
(finer-grained than the blanket WithSkipTLSHostVerify).
The knobs live on a non-exported config.KernelExperimentalConfig off
config.Config (not UserConfig), so they stay off the stable DSN surface
(mirroring Node's InternalConnectionOptions / Python's underscore kwargs).
The Thrift path rejects a non-nil block loudly at connect rather than
silently ignoring it, so a caller who forgets WithUseKernel learns the
option had no effect. OpenSession forwards each to the kernel C ABI via a
byte-buffer helper (cBytes); a reflective guard
(TestKernelExperimentalFieldsClassified) keeps a new field from slipping
either path unclassified.
mTLS client cert/key (needs an absent kernel C-ABI setter, K5) and the
CloudFetch on/off toggle (K3) are deliberately out of scope here — they are
tracked separately (PECOBLR-3652 / PECOBLR-3653).
Closes PECOBLR-3651.
Co-authored-by: Isaac
Signed-off-by: Mani Kaustubh Mathur <mani.mathur@databricks.com>
d0c999e to
2caa354
Compare
89c9b36 to
b147d0e
Compare
There was a problem hiding this comment.
Summary
Solid, focused PR — minor suggestions.
The experimental-option idiom (KernelExperimental off Config, WithKernel* prefix, fail-loud on Thrift) is the right design and consistent with the rest of the kernel backend work. cBytes / applyKernelTLS / trySetKernelTLS follow existing patterns well, and the reflective field-classification guard is a nice maintenance hook.
Main gap: no end-to-end test that Connect() actually rejects on the Thrift path (see inline on kernel_experimental_test.go). Everything else is polish / nice-to-have.
| // learns the option had no effect instead of connecting with a | ||
| // weaker-than-intended (or unconfigured) TLS trust store. | ||
| if c.cfg.KernelExperimental != nil { | ||
| return nil, errors.New("databricks: the WithKernel* options " + |
There was a problem hiding this comment.
nit (optional): programmatic sentinel
Kernel rejections wrap ErrNotSupportedByKernel so callers can fall back without matching message text. This Thrift-side rejection uses a plain errors.New.
Probably fine for experimental options, but if you expect kernel-first / Thrift-fallback callers, consider a symmetric sentinel (e.g. ErrRequiresKernelBackend) wrapped with %w here — same detection pattern as doc.go documents for the kernel path.
| // EXPERIMENTAL, kernel-only: the default (Thrift) backend rejects this at connect. | ||
| func WithKernelTrustedCerts(pem []byte) ConnOption { | ||
| return func(c *config.Config) { | ||
| kernelExperimental(c).TLSTrustedCertsPEM = pem |
There was a problem hiding this comment.
nit: slice aliasing
DeepCopy defensively copies the PEM bytes for pooled connections, but this stores the caller's slice by reference. If the caller mutates pem between NewConnector and Connect, behavior can change.
Consider copying on set to match DeepCopy:
if len(pem) > 0 {
kernelExperimental(c).TLSTrustedCertsPEM = append([]byte(nil), pem...)
} else {
kernelExperimental(c).TLSTrustedCertsPEM = pem
}(Separately: WithKernelTrustedCerts(nil) / []byte{} still allocates a non-nil KernelExperimental, so Thrift connect fails even though applyKernelTLS is a no-op — edge case, but worth being aware of.)
| // WithSkipTLSHostVerify, which relaxes both chain and hostname checks. | ||
| // | ||
| // EXPERIMENTAL, kernel-only: the default (Thrift) backend rejects this at connect. | ||
| func WithKernelSkipHostnameVerify() ConnOption { |
There was a problem hiding this comment.
nit: security warning consistency
WithSkipTLSHostVerify carries an explicit MITM warning. This option weakens TLS too (hostname only, but still). A one-liner WARNING: comment here would help callers understand the trade-off, similar to the existing option above.
| // branch is what rejects it. We assert the option→config wiring here (a non-nil | ||
| // KernelExperimental after applying a WithKernel* option is the signal the Thrift | ||
| // branch keys off). | ||
| func TestWithKernelTLSOptionsSetExperimental(t *testing.T) { |
There was a problem hiding this comment.
suggestion: add end-to-end Thrift rejection test
The file comment says this covers the "Thrift fail-loud signal", but this test only asserts option → config wiring. It doesn't call Connect() on the Thrift path.
TestKernelBackendNotCompiledIn is the model — a small untagged test like:
func TestWithKernelOptionsRejectedOnThriftPath(t *testing.T) {
c, err := NewConnector(
WithServerHostname("example.cloud.databricks.com"),
WithPort(443),
WithHTTPPath("/sql/1.0/endpoints/abc"),
WithAccessToken("token"),
WithKernelTrustedCerts([]byte("ca")),
)
if err != nil { t.Fatal(err) }
_, err = c.Connect(context.Background())
if err == nil {
t.Fatal("expected Connect to reject WithKernel* on Thrift path")
}
}would lock in the behavior the PR description emphasizes.
Context for reviewers
Part of the SEA-via-kernel backend for the Go driver: an alternative to the Thrift backend that talks to Databricks SQL through the Rust
databricks-sql-kernelstatic library over a C ABI (built into the driver behind thedatabricks_kernelbuild tag; the default pure-Go build is unaffected). This PR adds two TLS knobs to that backend.Stack position (each stacked on the one below):
How to read the diff: the new public surface is the two
With…options inconnector.go+ theirconfigfields; the interesting logic is (a) the Thrift-path reject and (b) theOpenSessionforwarding via the newcBytescgo helper. Everything else is wiring + tests.What
Exposes two kernel-only TLS options via an experimental-option idiom — no kernel change (both C-ABI setters already exist on kernel main, verified against the
origin/mainkernel header):WithKernelTrustedCerts(pem []byte)→kernel_session_config_set_tls_trusted_certs: adds a PEM CA bundle to the kernel's trust store on top of the system roots (corporate re-signing proxy / on-prem CA). Required rather than relying onSSL_CERT_FILE, which the kernel's rustls stack ignores.WithKernelSkipHostnameVerify()→kernel_session_config_set_tls_skip_hostname_verification: skips only the certificate hostname check while keeping chain validation (finer-grained than the blanketWithSkipTLSHostVerify).Closes PECOBLR-3651.
Design
config.KernelExperimentalConfigoffconfig.Config(NOTUserConfig), so they stay off the stable DSN/exported surface — same treatment asTLSConfig, mirroring Node's non-exportedInternalConnectionOptions.DeepCopycopies the CA byte slice.WithUseKernellearns the option had no effect rather than connecting with a weaker-than-intended trust store.OpenSessionforwards each to the kernel C ABI via a new byte-buffer helper (cBytes, mirroringcStr). A reflective guard (TestKernelExperimentalFieldsClassified) fails if a new experimental field isn't classified forwarded/rejected — so a future option can't silently be neither forwarded nor rejected.Scope boundary (deliberately excluded, deferred)
Bundle 1b only (the no-kernel-dependency part of the richer-TLS work). The following were confirmed absent from kernel main, so they need a kernel C-ABI setter first and are correctly deferred:
set_tls_client_certificate) → PECOBLR-3652 (K5).set_cloudfetch_enabled) → PECOBLR-3653 (K3).Testing
CGO_ENABLED=0: build + vet +go test ./...— 24 pkg ok, 0 fail.CGO_ENABLED=1 -tags databricks_kernel(linked against a locally-built kernel.a): build + vet + test — 24 pkg ok, 0 fail. NewTestSetKernelTLSdrives the real cgo setters; untaggedkernel_experimental_test.gocovers the option→config wiring, the classification guard, andDeepCopy.[]bytenormalization, and the idempotent double hostname-skip call when bothWithSkipTLSHostVerify+WithKernelSkipHostnameVerifyare set.Stacking / merge order
Stacked on #399 (
mani/sea-kernel-consolidated). NoKERNEL_REVchange here (inherits #399's pin). #401 (kernel logging) is stacked on top of this.This pull request and its description were written by Isaac.